Closed vateye closed 1 year ago
Hi, nice work to EVA-2.0, is there any plan to try EVA 2.0 training on CLIP or scaling EVA 2.0 to giant size?
Hi @vateye, all EVA-CLIP models (from 150M to 4.7B params) will be released in one week, please stay tuned.
Is there any information for 4.7B model? I am excited about it.
Hi @vateye, we have released all EVA-CLIP models and inference code at https://github.com/baaivision/EVA/tree/master/EVA-CLIP#model-card.
We offer performant CLIP models at all scales, range from 150M to 4.7B. Try them out if you are interested.
Thanks for your response. Is there any technical report for it?
Thanks for your response. Is there any technical report for it?
The technical report are coming very soon.
Hi, can you explain why Rope and subln is not used for EVA-2-e (CLIP)?
Hi, can you explain why Rope and subln is not used for EVA-2-e (CLIP)?
Because the 4.7B EVA-CLIP-E were trained before we started to explore RoPE, SwiGLU and sub-LN, etc.
We still call it EVA-02-CLIP-E because of the MIM teacher is EVA-01-CLIP-g.
Thanks for your response. Is there any technical report for it?
The technical report are coming very soon.
Hi, is there any update for EVA-CLIP report?
Thanks for your response. Is there any technical report for it?
The technical report are coming very soon.
Hi, is there any update for EVA-CLIP report?
Coming in 24h.
The report is available at https://arxiv.org/abs/2303.15389
Hi, nice work to EVA-2.0, is there any plan to try EVA 2.0 training on CLIP or scaling EVA 2.0 to giant size?